[quicwg/base-drafts] Server's SETTINGS can be omitted in 0-RTT (#1809)

[Mike Bishop <[e-mail]>]

#1808 adds a requirement to error out if the first frame on the control stream is not SETTINGS.  There's an existing error for duplicate control streams.  But what if the control stream never opens at all?

The current state of SETTINGS is that:
- In 1-RTT: Clients MUST wait until they see the server's SETTINGS frame before sending requests; the server MUST see the client's SETTINGS before sending any responses.  If either side fails to open the control stream and send SETTINGS, the connection stalls.
- In 0-RTT:  Servers MUST wait until they see the client's SETTINGS frame before sending responses; this is no different.  Clients, however, assume previous settings and update them *when* the server's SETTINGS frame is eventually seen.

It is a violation for the server to send responses without having sent the SETTINGS frame.  However, there is no requirement that clients validate this behavior because of reordering.  This means that a spec-conformant client would interop just fine with a server that never sends SETTINGS on 0-RTT connections.

Should we introduce a wait-for-SETTINGS requirement later in the connection (for example, before sending new requests in 1-RTT), or just hope that it's easier for servers to do the same compliant thing for both 0-RTT and 1-RTT connections?

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/1809